The mobility of the future is autonomous: people and machines share the traffic space. They interact and cooperate. However, the necessary level of cognitive skills of autonomous road users can only be achieved through a comprehensive use of machine learning methods.

With regard to the scalability, transferability and traceability of the developed data-based driving functionality, hurdles remain on the way to higher degrees of autonomy, nevertheless. These limitations stem from the current system architecture and the specific discriminative machine learning methods employed.

Generative AI offers an alternative approach and has impressively demonstrated its capabilities and technological maturity in recent applications such as large language models or text-to-image generators. By using generative methods and the resulting development of foundation models, nxtAIM tries to initiate nothing less than a paradigm shift in system architecture and AI methodology.

Today’s system architectures are based on linear, unidirectional information processing along the so-called chain of effects consisting of perception, environment modeling, prediction & planning and subsequent implementation. Generative methods are used to create a feedback channel and information processing is expanded bidirectionally.